Bidirectional LSTM-based compensation of potential-asymmetry induced errors in double probe plasma diagnostics

Abstract The double Langmuir probe serves as a fundamental diagnostic tool for plasma characterization, providing critical insights into plasma parameters. The conventional double probe theory relies on idealized assumptions of identical probe geometry and equipotential conditions. However, experimentally obtained I – V characteristics often exhibit significant deviations, manifesting as curve offsets from the origin and asymmetric features. Direct application of conventional fitting methods to such non-ideal measurement data introduces diagnostic errors due to the violation of these assumptions. Given the limitations imposed by manufacturing tolerances and complex experimental conditions, neither experimental refinements nor theoretical corrections alone can adequately resolve this issue. To address the diagnostic errors induced by potential asymmetry in double-probe measurements, this study proposes a machine learning method by automatically extracting features from I – V characteristics to learn plasma diagnostic patterns. We established a DC glow discharge experimental platform to systematically acquire I – V characteristics under various discharge voltage and gas pressure, while intentionally introducing potential differences through precise angular displacement of the probes. This process generated a comprehensive dataset incorporating potential asymmetry effects. A bidirectional long short-term memory network was developed, utilizing probe current as input features and electron temperature and density calculated under symmetric potential conditions as labels. The model effectively reduces errors induced by potential asymmetry in double-probe measurements, achieving 97% accuracy in electron temperature and density prediction. This study provides a novel approach to enhancing the precision and efficiency of double Langmuir probe plasma diagnostics in complex environments.